The behavior of an NB-IoT communications link using DRL-based dynamic channel selection is explored in the presence of a DRL-based jamming attacker. A reinforcement learning framework is developed using PPO agents, and their behavior is examined under different channel selection conditions. A Markov Decision Process (MDP) is used to model the channel evolution, and a pre-trained victim agent is deployed alongside a pre-trained jammer. Performance results are compared with and without a victim agent defense strategy. Overall, the defensive strategy employed was able to evade the jammer, but not effective enough to be a viable defensive strategy.
*Requires Reinforcement Learning and LTE Toolboxes
-
To train the initial victim agent, run 'pretrain_victim_script.m'
-
To train the jammer agent, run 'pretrain_jammer_script.m'
-
To train the defensive victim agent, run 'victim_defense_script'
-
To run NB-IOT simulation, run 'main_script'
- Adjust 'doSim' to control whether NB-IOT sim runs
- Adjust 'numTrials' to control the number of time steps
- Adjust 'run_with_jammer' to control whether the jammer is present
- Adjust 'run_with_fortified_victim' to control whether the initial victim or defensive victim is used